ci: Enforce conventional commit format in PR title

[augustebaum]

• Add commitlint workflow to lint the PR title (not the individual PR commits)
• Make sentence-case the only allowed PR title format (first letter must be capitalized, like "Merge commit ..." or "Revert commit ...")

Closes #66

[tuscland]

I love this! Can we mention expected rules in the contributing guide?

[augustebaum]

He can therefore easily override this PR title rule.

This is a good point; I will do some research to see how to deal with this. In the meantime, only people with write access to the repo (i.e. the Probabl product team) can do this when merging a PR, so as long we're minimally disciplined I don't think that's an issue

[augustebaum]

it should be done at the commits level.

You mean check individual commits inside a PR? I don't mind, it just heavily increases dev friction and in the end all the commits are squashed

[augustebaum]

it's premature

This is a pre-requisite for proper release management, e.g. for updating the changelog automatically. It's a simple change, and it will immediately make our main branch cleaner.

[thomass-dev]

He can therefore easily override this PR title rule.

This is a good point; I will do some research to see how to deal with this. In the meantime, only people with write access to the repo (i.e. the Probabl product team) can do this when merging a PR, so as long we're minimally disciplined I don't think that's an issue

Time for joke: if we're disciplined, we don't need such a test, since only the maintainers can merge a PR and therefore edit the merge-commit message and respect the convention :) .

I'm truly convinced that if we want to base an automatic release management on conventional commits, we need to add some strong rules that can't be ignored. To err is human.

---

I don't know if its possible to run a workflow after squashing but before merging.

[tuscland]

I agree with @thomass-dev. We can start by describing good conventions, exercise them manually and then implement them automatically.

e.g. for updating the changelog automatically

I am not sure. I don't like changelogs based solely on automatically generated statements; they should primarily have an editorial perspective.

[augustebaum]

Time for joke: if we're disciplined, we don't need such a test, since only the maintainers can merge a PR and therefore edit the merge-commit message and respect the convention :) .

Obviously, you need much less discipline to not change a green PR message, than to make sure every PR you review abides by a full convention. This is a net positive.

We can start by describing good conventions, exercise them manually and then implement them automatically.

I don't think this convention will be strictly followed unless there are automatic checks for it; furthermore, I don't want to do a job that this PR could easily achieve.

I don't like changelogs based solely on automatically generated statements.

Point taken; taking the list of commits is a first step, that can then be used to write up more digestible updates.